Fix stdatomic case in rbimpl_atomic_u64_fetch_add

On some platoforms, 64bit atomic operations need the dedicated helper
library.

Author: nobu

configure.ac

@@ -1746,6 +1746,7 @@ AS_IF([test "$GCC" = yes], [
 		    [rb_cv_gcc_atomic_builtins=no])])
     AS_IF([test "$rb_cv_gcc_atomic_builtins" = yes], [
 	AC_DEFINE(HAVE_GCC_ATOMIC_BUILTINS)
+	AC_CHECK_LIB([atomic], [__atomic_fetch_add_8])
 	AC_CACHE_CHECK([for 64bit __atomic builtins], [rb_cv_gcc_atomic_builtins_64], [
 	    AC_LINK_IFELSE([AC_LANG_PROGRAM([[@%:@include <stdint.h>
 			uint64_t atomic_var;]],

ruby_atomic.h

@@ -2,6 +2,9 @@
 #define INTERNAL_ATOMIC_H
 
 #include "ruby/atomic.h"
+#ifdef HAVE_STDATOMIC_H
+# include <stdatomic.h>
+#endif
 
 #define RUBY_ATOMIC_VALUE_LOAD(x) rbimpl_atomic_value_load(&(x), RBIMPL_ATOMIC_SEQ_CST)
 
@@ -76,9 +79,9 @@ rbimpl_atomic_u64_fetch_add(volatile rbimpl_atomic_uint64_t *ptr, uint64_t val)
     return InterlockedExchangeAdd64((volatile LONG64 *)ptr, val);
 #elif defined(__sun) && defined(HAVE_ATOMIC_H) && (defined(_LP64) || defined(_I32LPx))
     return atomic_add_64_nv(ptr, val) - val;
+#elif defined(HAVE_STDATOMIC_H)
+    return atomic_fetch_add_explicit((_Atomic uint64_t *)ptr, val, memory_order_seq_cst);
 #else
-    // TODO: stdatomic
-
     // Fallback using mutex for platforms without 64-bit atomics
     static rb_native_mutex_t lock = RB_NATIVE_MUTEX_INITIALIZER;
     rb_native_mutex_lock(&lock);